On Wed, 4 Feb 2004 19:10:56 -0800 mcq1@viz.cacr.caltech.edu wrote:
I just noticed a 1.02 BIOS go up in the Tyan directory. You might want to wait a bit before trying it out, since it seems to have re-broken DMA to the 3ware card and eaten my / again...
Was this with the latest update kernel ? And do you have more than 3GB of RAM?
This was both after boot and during POST (corrupt card ID messages and system hang), so it's definately a hardware thing. Tyan was quite helpful and pointed me at this:
Ok thanks. Please always mention such things when you talk about corruptions.
And the 3ware iommu bug also hit Qlogic cards, yes? I can't believe those
Yes it did. But I finally gave in and disabled the optimization to avoid corrupting people's disks. I still believe it's a bug somewhere in hardware and not the software though that is probably just hidden by the more frequent flushes.
have signalling problems considering they plug into every weirdo PCI-carrying not-a-PC jumbo datacenter rack monster on the planet without a problem.
I don't consider it that unlikely that two independent card have the same issues. Maybe they used the same macro cell to interface to PCI bought from some IP vendor or whatever. Or maybe it's a bug in the chipset. Or a combination of both.
But recall that the iommu bug (as far as I know) was never explained -- it was merely noted that the flush optimization triggered it, so the optimization was backed out. So it might have been the same tim'rous signalling beastie aggravated in both cases...er, I guess.
Yep.
The upshot at the moment seems to be that if Tyan hears more complaints from more people for more cards than just the 3ware, perhaps we'll here more.
Thanks for the update. Unfortunately we hear the complaints too and people blame such things on Linux :-/ But it's good that it at least breaks at POST now too, this makes it clearer where the problems are. -Andi